Why Matched Sets Matter on Multi-Belt Drives

Why this matters

"Replace the whole set" is one of the few belt rules everybody has heard and almost nobody can defend, which is why it gets ignored on a Friday afternoon when one belt of three has let go and there is one belt on the truck. The defence is not tradition. Belts on a multi-groove drive share load in proportion to their own tension, their tension is set by their own length relative to the others, and the length differences that matter are a few tenths of a percent, which is invisible with a tape and is most of the stretch the drive was installed with. Get it wrong and the drive does not fail; it fails three times, in sequence, and every one of those failures gets written up as a bad belt.

Isolate, and clear the wreckage properly

Lock out and tag the motor at its disconnect. A wheel holds rotational energy after the power is off, which is stored mechanical energy under 29 CFR 1910.147, so verify zero rotation by eye and block the wheel or close and secure the damper before reaching past the guard. Where one belt of a set has already failed, expect fragments and cord whiskers still wrapped in the drive and inside the guard: clear them locked out, with cut-resistant gloves, and vacuum the rubber dust with a HEPA-filtered vacuum rather than blowing it into the room. Slacken the drive fully to remove or fit belts; prying one over a rim with a screwdriver cuts the tensile cords and can snap belt and tool back at you. Any running observation is taken from outside a closed guard and out of the plane of rotation, since a belt that lets go at speed leaves along that plane. The guard goes back on before restart under 29 CFR 1910.219 in general industry or 29 CFR 1926.300(b) on a construction site.

The gate

Measure the deflection force on every belt in the set, at the same span length and the same deflection distance, with the drive at rest. If the spread between the highest and lowest reading exceeds 15 percent of the set's mean, the set is mismatched and every belt in it gets replaced. At or below 15 percent, the belts will share load and a single replacement belt may stand.

The unit of analysis is per belt, on a per-drive set, measured at the same span. It is one gate on the spread, not a pair of conditions, and the action it triggers is all or nothing: you do not correct a mismatched set by adjusting tension, because tension is the thing you are measuring and every belt on the drive shares one center distance.

Where the 15 percent comes from: deflection force is proportional to strand tension, and load shares roughly in proportion to tension, so on a three-belt drive a 15 percent high-to-low spread leaves the most-loaded belt about 7 percent above its fair share, which is inside the reserve a service-factored drive selection already carries. Treat it as a working rule and use your maker's figure where one exists; most do not publish one, which is why this needs deriving rather than quoting.

Why a fraction of a percent of length is a big number

Every belt on the drive is stretched over the same two sheaves at the same center distance, so they all end up at the same installed length. What differs is what each one started at.

A belt's tension comes from its strain, meaning how far it has been stretched relative to its own free length. Treating the belt as a linear spring in tension, and ignoring the small relief that sheave and shaft deflection provide, if the free lengths differ by a fraction delta and the mean installed strain is e0, then the two belts sit at strains of about e0 plus delta/2 and e0 minus delta/2, and their tensions are in that same ratio.

That is the whole trick, and it is why the numbers feel wrong the first time. The mismatch is not measured against belt length. It is measured against installed stretch, which is small. Take an illustrative installed strain of 0.5 percent, and a free-length difference of 0.4 percent between two belts, which on a 60 in belt is under a quarter of an inch and undetectable by eye on the machine. The strains become 0.7 percent and 0.3 percent, and the tension ratio is 0.7 divided by 0.3, about 2.3 to 1. One belt is carrying more than twice what its neighbour carries.

Two consequences follow that people get wrong in the field:

An automatic or spring-loaded tensioner does not fix mismatch. It sets the mean tension by floating the center distance, and all the belts share that one center distance, so the spread between them is untouched. It changes the average and not the split.

Stiffness matters as much as length. Tension is strain times the belt's longitudinal stiffness, so two belts of identical length from different makers or different constructions still split the load unevenly. Never mix makers, constructions or sections in a drive even when the nominal lengths agree.

What "matched" actually means

Not "same part number." A part number gives a nominal length; matching is about the tolerance band around it. Historically that meant belts were measured under a standard tension and packaged in sets carrying a matching code, and you bought the set.

Modern belt construction holds length tolerance tighter than it once did, and several makers now state that any two belts of the same size, construction and make will fall inside their matching band, so matched sets are no longer sold separately. That is a real change and it is a per-maker claim, so read theirs rather than assuming it: check whether the claim covers belts from different production lots, and whether it applies to the construction you are fitting. What it never covers is mixing a new belt with used ones, because a used belt's length has changed for reasons no manufacturing tolerance controls.

Banded belts answer this differently, by joining the belts into one part. They cannot mismatch and they resist the whipping that shock loads induce in individual belts, at the cost of the whole band being one replacement.

One gate, two drives, opposite answers

Drive A. Three-groove blower drive, four years in service, one belt thrown. A new belt of the correct size and construction fitted alongside the two survivors, tension set, then all three measured at the same span and deflection: new belt 9.5 lb, survivors 4.2 lb and 4.0 lb.

Mean is 5.9 lb. High-to-low spread is 5.5 lb, which is 93 percent of the mean, far outside the gate. The new belt against the survivors' 4.1 lb average is a ratio of 2.3 to 1, which is where the length arithmetic above said it would land: four years of permanent set plus sidewall wear letting the old belts ride deeper has left them effectively longer than the belt they were made the same as.

Its share of the load is 9.5 divided by 17.7, or 54 percent, against a fair share of 33 percent. It is carrying 1.6 times its design load, on a drive that was selected assuming three belts pulling equally. It will fail well inside the interval anyone expects, and when it does, the two survivors have to carry the whole load at 50 percent each, or 1.5 times their design share. They go next, and the shop books three belt failures and a reputation for bad belts.

Verdict: replace all three.

Drive B. Same drive design, commissioned six hours ago. One belt was cut during installation when it was levered over the rim with a screwdriver, which is its own lesson. One new belt of the same size, maker and construction fitted, tension set, all three measured: 6.8 lb, 6.6 lb, 6.5 lb.

Mean 6.63 lb, high-to-low spread 0.3 lb, about 5 percent of the mean. Inside the gate. Six hours is enough for some seating but not enough for the differential permanent set that separated Drive A's belts, and the measurement says so.

Verdict: the single belt stands. Re-check the spread at the run-in re-tension, because if seating separates them the gate will catch it then.

What the two cases have in common. Identical drive, identical event, opposite answers, and the thing that decided it was neither age nor policy but three readings that take about four minutes with a tester already in the bag. Age is the prior, not the gate. A drive that has run four years might read tight if the belts were matched and lightly loaded, and a drive commissioned last week can read wide if somebody fitted a belt from a different maker.

What changes the answer

Groove wear does not respect a set. If one groove has dished further than the others, that belt rides deeper and reads low no matter what its length is, and replacing belts will not correct it. Gauge the grooves before you conclude the belts are mismatched, because the readings look the same and the fixes are not.

Very long belts widen everything. Manufacturing tolerance is expressed as a proportion of length, so on a long-center drive the absolute length spread is larger while the installed strain is not, which makes the tension spread worse for the same tolerance class. Long-center multi-belt drives are the ones most worth buying as a set.

A two-belt drive can hide it. With three or more belts the spread is obvious across the readings. With two, both could be off in the same direction from the drive's requirement and still read close to each other, so check the readings against the maker's table as well as against each other.

How to verify you got this right

• Measure every belt, not the set. A single reading on a multi-belt drive tells you about one belt and nothing about the split. Same span, same deflection distance, all of them.
• Write down all the readings, not the average. The spread is the finding, and next visit's spread compared against this one is how you see a set separating before it cascades.
• Re-measure at the run-in re-tension. New belts seat, and a set that read even on day one can separate as they bed in. This is the last cheap chance to catch it.
• Label the drive with the belt count, size and construction fitted. Half of all mismatch arrives as somebody's honest attempt to match what was in there, from a photograph of a worn part number.

References

• 29 CFR 1910.147 for mechanical isolation and stored rotational energy; 29 CFR 1910.219 (general industry) and 29 CFR 1926.300(b) (construction) for guarding of belts, pulleys and sheaves
• Belt manufacturer engineering data for length tolerance and matching classes, statements on set-free or interchangeable construction, and banded belt selection
• See related: How to Tension a Belt Without Guessing; What a Worn Sheave Groove Does to a New Belt; What Belt Tension Actually Controls